Peyman Adibi, a faculty member of Artificial Intelligence Department, Faculty of Computer Engineering, University of Isfahan, Iran will give a talk on
“Multiple kernel learning and linear manifold topographic maps for supervised dimensionality reduction”
on friday february 2nd, 2pm, in room Chartreuse at GIPSA-lab/DIS, Grenoble.
An abstract of the talk is given below:
In this seminar, two parametric dimensionality reduction approaches are introduced. The first one (TS-MKL-SDR) is a nonlinear supervised method and the second (LMTM) is a local linear unsupervised model.
TS-MKL-SDR: Two-stage multiple kernel learning for supervised dimensionality reduction
In supervised dimensionality reduction methods for pattern recognition tasks, the information of the class labels is considered through the process of reducing the input dimensionality, to improve the classification accuracy. Using nonlinear mappings for this purpose makes these models more appropriate for nonlinearly distributed data. TS-MKL-SDR is a new nonlinear supervised dimensionality reduction model, based on multiple kernel learning paradigm. In the first stage, three suitable criteria for supervised dimensionality reduction are used to find the base kernel weights to make a new suitable kernel. In the second stage, the kernel discriminant analysis method is employed for nonlinear supervised dimensionality reduction using the kernel computed in the first stage. Many experiments on a variety of real-world datasets including handwritten digits images, objects images, and other datasets, show that the proposed approach among a number of well-known related techniques, results in accurate and fast classifications.
LMTM: Linear manifold topographic maps
The model uses unsupervised algorithms for learning the underlying regional linear data manifolds in a neural topographic map. An on-line approach based on a gradient descent process, and a batch algorithm based on an expectation-maximization procedure applied on the map energy function, are discussed. Then, a regional dimension learning process devised for the model is described. Learning different on- and off-manifold extents is also considered. Experimental results are finally reported.